Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/79261Full metadata record
| ???org.dspace.app.webui.jsptag.ItemTag.dcfield??? | Value | Language |
|---|---|---|
| dc.contributor.advisor | 莊永裕(Yung-Yu Chuang) | |
| dc.contributor.author | Chun-Wei Chen | en |
| dc.contributor.author | 陳俊瑋 | zh_TW |
| dc.date.accessioned | 2022-11-23T08:56:54Z | - |
| dc.date.available | 2022-01-17 | |
| dc.date.available | 2022-11-23T08:56:54Z | - |
| dc.date.copyright | 2022-01-17 | |
| dc.date.issued | 2022 | |
| dc.date.submitted | 2022-01-10 | |
| dc.identifier.citation | [1] HE, Kaiming, et al. Mask r-cnn. In: Proceedings of the IEEE international conference on computer vision. 2017. p. 2961-2969. [2] SHEN, Xing, et al. Dct-mask: Discrete cosine transform mask representation for instance segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021. p. 8720-8729. [3] ZHANG, Gang, et al. RefineMask: Towards High-Quality Instance Segmentation with Fine-Grained Features. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021. p. 6861-6869. [4] CHEN, Hao, et al. BlendMask: Top-down meets bottom-up for instance segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020. p. 8573-8581. [5] CAO, Jiale, et al. Sipmask: Spatial information preservation for fast image and video instance segmentation. In: Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part XIV 16. Springer International Publishing, 2020. p. 1-18. [6] WANG, Chien-Yao; BOCHKOVSKIY, Alexey; LIAO, Hong-Yuan Mark. Scaled-yolov4: Scaling cross stage partial network. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021. p. 13029-13038. [7] BOLYA, Daniel, et al. Yolact: Real-time instance segmentation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019. p. 9157-9166. [8] REN, Shaoqing, et al. Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 2015, 28: 91-99. [9] TIAN, Zhi, et al. Fcos: Fully convolutional one-stage object detection. In: Proceedings of the IEEE/CVF international conference on computer vision. 2019. p. 9627-9636. [10] GE, Zheng, et al. OTA: Optimal Transport Assignment for Object Detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2021. p. 303-312. [11] LIU, Shu, et al. Path aggregation network for instance segmentation. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2018. p. 8759-8768. [12] ZHENG, Zhaohui, et al. Enhancing geometric factors in model learning and inference for object detection and instance segmentation. IEEE Transactions on Cybernetics, 2021. [13] LI, Yi, et al. Fully convolutional instance-aware semantic segmentation. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2017. p. 2359-2367. [14] FU, Cheng-Yang; SHVETS, Mykhailo; BERG, Alexander C. RetinaMask: Learning to predict masks improves state-of-the-art single-shot detection for free. arXiv preprint arXiv:1901.03353, 2019. [15] CHEN, Kai, et al. Hybrid task cascade for instance segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019. p. 4974-4983. [16] CHEN, Xinlei, et al. Tensormask: A foundation for dense object segmentation. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 2019. p. 2061-2069. | |
| dc.identifier.uri | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/79261 | - |
| dc.description.abstract | 本研究提出一個即時的實例分割系統，實例分割的目的在於找出圖片中的各個物件，並區分出每個實例之間的差異，例如圖中有兩個人，就需要區分出是兩個不同的人。目前有許多基於Mask R-CNN的two-stage方法，能達到相當高的MAP，例如RefineMask可以於coco2017val達到45.3 MAP，但無法達到Real-time。同時Real-time的作法，例如BlendMask與SipMask MAP則僅有36.3與34.2。本系統基於scaled yolov4，並將其延伸，在維持其原本Real-time的特性下，加入了實例分割的功能，實例分割的部分主要參考BlendMask，並加上了錨定（anchor）的概念。同時於NMS提出了一個創新的做法，Mask confidence。並於FPN的P2加入一個Saliency Map branch進行self-supervised learning。最終可在Real-time的情況下，於coco2017val達到42.6 MAP。 | zh_TW |
| dc.description.provenance | Made available in DSpace on 2022-11-23T08:56:54Z (GMT). No. of bitstreams: 1 U0001-1001202210084600.pdf: 1735583 bytes, checksum: 323681cb32b93c6e2ab107af513e2b20 (MD5) Previous issue date: 2022 | en |
| dc.description.tableofcontents | 致謝 i 摘要 ii Abstract iii Content iv List of Figures vi List of Tables vii 1 Introduction 1 1.1 Motivation 1 1.2 Objective: Real Time Instance Segmenation 1 1.3 Contributions 2 1.4 Chapter Outline 2 2 Related Works 3 3 Data Augmentation 5 3.1 Mosaic 5 3.2 Perspective Warping and Flipping 5 4 Network Architecture 7 4.1 Object Detection: Yolov4 7 4.2 Detection Heads 9 4.3 Bases Branch and Blender 10 4.4 Auxiliary Saliency Branch 11 4.5 Whole Network Architecture 13 5 Label Assignment 14 5.1 Label Assignment in Yolov4 and OTA 14 5.2 Strictly Optimal Transport Assignment 16 6 Mask Confidence 16 7 Experiments 17 7.1 Results 17 7.2 Comparison with other works 19 7.3 Number of Neighbors in OTA 20 7.4 Anchor for Mask 20 7.5 Ablation Study 21 8 Conclusions 23 Reference 24 | |
| dc.language.iso | en | |
| dc.title | 基於錨定的即時實例分割系統 | zh_TW |
| dc.title | Anchor-based real-time instance segmentation system | en |
| dc.date.schoolyear | 110-1 | |
| dc.description.degree | 碩士 | |
| dc.contributor.oralexamcommittee | 廖弘源(Keh-Ming Shyue),林永隆(Yu-Chen Shu),王建堯 | |
| dc.subject.keyword | 電腦視覺,物件偵測,實例分割,深度學習,錨定, | zh_TW |
| dc.subject.keyword | Computer vision,Object detection,Instance segmentation,Deep learning,Anchor-based, | en |
| dc.relation.page | 27 | |
| dc.identifier.doi | 10.6342/NTU202200031 | |
| dc.rights.note | 同意授權(全球公開) | |
| dc.date.accepted | 2022-01-11 | |
| dc.contributor.author-college | 電機資訊學院 | zh_TW |
| dc.contributor.author-dept | 資訊工程學研究所 | zh_TW |
| Appears in Collections: | 資訊工程學系 | |
Files in This Item:
| File | Size | Format | |
|---|---|---|---|
| U0001-1001202210084600.pdf | 1.69 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.